7. #GlobalAICommun
ity
#GlobalAIBackTogeth
er
• By using Machine Learning, Netflix studies their users’ behavior and
viewing history in order to recommend the “right” artwork for them.
• If the artwork representing a title captures something compelling to
users, then it acts as a gateway into that title and gives subscribers
some visual “evidence” for why the title might be good for them.
10. #GlobalAICommun
ity
#GlobalAIBackTogeth
er
Proven ROI
Customer
Experience
Transformation
Pays Off
Source: McKinsey, Microsoft Services
Before After
Increase Customer
Satisfaction by up to
20 %
Before After
Increase Revenue
Growth by
10-15 %
Before After
Lower Cost To
Serve by
15 a 20 %
Brands that excel at personalization deliver 5-8 times the marketing ROI
and boost their sales by more than 10% over companies that don’t
personalize
11. #GlobalAICommun
ity
#GlobalAIBackTogeth
er
Microsoft Experience and Results
Xbox Home
RESULTS
+40% lift in engagement for items
+2.5% increase in mixer engagements
MSN News
RESULTS
+25% improvement in News clickthrough
Personalized: News content on
top of page in MSN.com
Reward: Click on content on the
first slot
Model updated every 5 minutes
Personalized: Type
of content in hero
position, item in
secondary place.
Reward: Click and
engagement
14. #GlobalAICommun
ity
#GlobalAIBackTogeth
er
Reinforcement Learning
• RL refers to a branch of AI / ML, which is aimed
at goal-oriented problems.
• RL algorithms can achieve complex goals by
maximizing a reward function in many steps,
for example, points won in a multi-step game.
• The reward function works in a similar way to
incentivizing a child with sweets and
punishments, with the algorithm penalizing
itself when it makes a wrong decision and
rewarding itself when it makes the right one;
this is a reinforcement.
15. #GlobalAICommun
ity
#GlobalAIBackTogeth
er
Reinforcement Learning in Recommendation
systems
• Recommenders: given a user profile and
categorized content, a system makes a
recommendation based on popularity,
interests, demographics, frequency, and other
features.
• RL's boost aspect allows it to more quickly
adapt to real-time changes in the user's
sentiment and profile, without the need for
explicit (re) training.
Enterprise adoption also appears to be
gaining momentum with the recent
availability of cloud APIs such as Azure
Personalizer and Google's RecSim.
20. #GlobalAICommun
ity
#GlobalAIBackTogeth
er
Azure Cognitive Services
Computer Vision
Face API
Custom Vision
Text-to-Speech
Speech-to-Text
Speech Translator
Speaker Recognition
Language Understanding
Text Translator
Text Analytics
QnA Maker
Anomaly Detector
Content Moderator
Personalizer
microsoft.com/cognitive
Vision Speech Language Decision
21. #GlobalAICommun
ity
#GlobalAIBackTogeth
er
Azure Applied AI Services
• Go live with AI solutions in days—not
months.
• Generate tangible value for your
organization quickly with AI services for
common business processes.
• Azure Applied AI Services bring together
Azure Cognitive Services, task-specific AI,
and business logic to offer you turnkey AI
services for common business processes.
25. #GlobalAICommun
ity
#GlobalAIBackTogeth
er
Azure Personalizer
• Microsoft is the first company to offer reinforcement learning for the
enterprise thanks to Personalizer service.
• The model’s learning cycle runs at digital speed and it learns from a simple
compensation score that is optimized to meet business goals.
• It allows to offer smarter experiences for each user that improve over time.
• Unlike recommendation engines that adopt some features and suggest items
from a large catalog, Personalizer takes some items with sophisticated features
and predicts the best result for the specific user on that device at the same time.
26. #GlobalAICommun
ity
#GlobalAIBackTogeth
er
Sample scenario
Have Fun!
…
We think you’ll like
?
Watch
Play
Join
Personalizer
Watch Action 2 Info
Action 3 Info
User & Context
Info
Action 1 info
Your Users
Your business
logic observing &
evaluating user
behavior
Reward Score
Reward (r)
Personalizer
Service
Rank (..)
Join
Play
Join
Watch
Play
Join
28. #GlobalAICommun
ity
#GlobalAIBackTogeth
er
Typical training now has new patterns
Ranking
Leverage existing recommendation engine data,
approved user profiles, or context data sources
Feedback
Discover new patterns and retrain on existing models,
improve frameworks over time
Product /
Media Catalog
e.g. 500,000 items
User
30 Items
Personalizer
Top contextual &
personalized
products
Recommendation Engine
Existing Personalization
Existing Segmentation,
Collaborative Filtering, Etc.
Editorial Team
Editorial Team creates content
Typical training now has new patterns
30. #GlobalAICommun
ity
#GlobalAIBackTogeth
er
• Send information (features) about your users and the content (actions) to
personalize.
• Personalizer responds with the top action.
• Send feedback to Personalizer about how well the ranking worked as a
number typically between 0 and 1.
44. #GlobalAICommun
ity
#GlobalAIBackTogeth
er
Where to use Personalizer?
• Build individual experiences: which article(s) are featured on a news
website for a user.
• Determine which recommended product should appear to each
customer in an online store.
• Suggest UI elements, such as filters, to apply to a photo.
47. #GlobalAICommun
ity
#GlobalAIBackTogeth
er
Do you want to learn more?
What is Personalizer?
https://azure.microsoft.com/en-us/services/cognitive-
services/personalizer/
Some Demos
https://personalizationdemo.azurewebsites.net/
https://personalizercontentdemo.azurewebsites.net
Quickstart
https://docs.microsoft.com/en-us/azure/cognitive-
services/personalizer/quickstart-personalizer-sdk
48. #GlobalAICommun
ity
#GlobalAIBackTogeth
er
Do you want to learn more?
Configure Personalizer learning loop
https://docs.microsoft.com/en-us/azure/cognitive-services/personalizer/how-to-settings
Tutorial: Add Personalizer to a .NET web app
https://docs.microsoft.com/en-us/azure/cognitive-services/personalizer/tutorial-use-
personalizer-web-app
Personalizer on Microsoft Learn
https://docs.microsoft.com/en-us/learn/modules/intro-to-personalizer/
Pluralsight
https://www.pluralsight.com/courses/microsoft-azure-cognitive-services-personalizer